prusek-spheroid 6.4

Spheroid segmentation package

Prusek-Spheroid

Prusek-Spheroid is a Python package designed for spheroid segmentation based on provided microscope images. This package provides an easy-to-use interface and functionalities that are essential for determination of properties and characteristics of the spheroids.

Installation

Installing Python

For Windows:

1. Download Python from python.org.
2. Run the downloaded installer. Ensure to check the "Add Python to PATH" option.
3. Verify the installation by opening CMD and typing python --version.

For MacOS/Linux:

1. Check if Python is installed by typing python3 --version in the terminal. If not installed:
   • On MacOS: Install via Homebrew with brew install python3.
   • On Linux: Install using sudo apt-get install python3.
2. Verify the installation by typing python3 --version in the terminal.

Installing Miniconda

1. Download Miniconda from the official Miniconda website.
2. Install Miniconda and follow the on-screen instructions.
3. Verify the installation by opening a new terminal or CMD and typing conda list.

Creating a Virtual Environment and Installing Prusek-Spheroid

1. Create a virtual environment using Miniconda: conda create -n myenv python=3.x. Replace x with python version 3. The recommended version of python 3 is 3.8.
2. Activate the virtual environment: conda activate myenv (Windows) or source activate myenv (MacOS/Linux).
3. Install the Prusek-Spheroid package: pip install prusek_spheroid.

Installing PyTorch

Prusek-Spheroid requires PyTorch for certain functionalities. Follow these steps to install PyTorch in your virtual environment:

1. Activate your virtual environment (if not already activated).
2. Install PyTorch using pip by running the following command: pip install torch

For specific versions and configurations (e.g., with CUDA support), visit the PyTorch official website for the appropriate installation command.

Running the Package

Activate your Conda virtual environment using the command: conda activate environment_name

Use the name of your created conda virtual environment instead of environment_name.

If you don't know the name, you can get it from the list of Conda virtual environments you've created using the command: conda info --envs

To use the package, first ensure it is up to date: pip install --upgrade prusek_spheroid

then run the program using the command: python -m prusek_spheroid.GUI

Introductory window

At the beginning, the user can choose which software he wants to use. There are two main ones to choose from. The first one is used to segment the spheroid images. It saves these segmentations and offers the possibility to determine the properties (quantify) of the spheroids based on these segmentations. The second option is used only to determine the properties of the spheroids based on the provided binary masks of the previously segmented spheroids. The other two ancillary software are image manipulation support programs. The first of these (left) is software for converting COCO 1.0 annotations to Masks, or the other way around. The second software is used to consolidate multiple image folders into one. The software assumes that the input directory branches into multiple directories and that the end folders contain images in BMP format. The software unifies these images from the end folders into one, changes their names to match the location in the directories, and removes duplicate images.

User Guide for Spheroid Segmentation GUI

Prusek-Spheroid is a sophisticated Python package equipped with a user-friendly graphical interface (GUI) that facilitates image segmentation and optimization tasks. This guide provides an overview of the GUI functionalities and how to use them effectively. The whole program is based on the knowledge of segmentations of several (approximately 15 to 20) spheroids. Based on these segmentations, the program learns to segment the remaining spheroid images in the dataset (project). So far the only possible formats in which the segmentations can be loaded are the COCO 1.0 format or loading masks and corresponding images. This is the format supported by the CVAT (Computer Vision Annotation Tool) platform.

Using the Spheroid Segmentation GUI

The Prusek-Spheroid GUI is designed to be intuitive, providing a range of functionalities for effective image segmentation. Here’s a detailed overview of the key elements:

Important: When using Prusek-Spheroid on Windows, folder names that are part of the addresses in your project must be unaccented.

1. Input File and Folder Selection: These two tabs are used for selecting and loading the annotations from CVAT in COCO 1.0 format (or loading masks and their corresponding images) and directory where your dataset images for segmentation are stored. You can easily navigate and choose the required folders and zip file (in COCO 1.0 format from downloaded from CVAT) that contain your image files.

Important: When downloading COCO 1.0 annotations from CVAT, it is necessary to check the "save with images" box to add the "images" folder to the resulting downloaded ZIP file in addition to the "annotations" folder.

Important: Supported image formats are PNG, JPG, JPEG, BMP, TIFF, TIF

2. Output Folder Selection: After processing, the results will be saved in the selected output folder. This feature allows you to specify where you want the segmented images and other outputs to be stored. If selected, the properties of the contours are also saved in the resulting folder as an excel file. Furthermore, the JSON file with the optimal parameters is also saved in the selected output folder in the "IoU" subfolder.

3. Project Name Field: Here, you enter the name of your project. This name will be used for organizing and saving the results in a structured manner.

4. 'I Already Know the Parameters' Checkbox: If you have pre-determined parameters from a previous Gradient descent project, you can use this checkbox. It's useful when you want to segment images using already optimized parameters without going through the Gradient descent process again. Is also recommended to use when knowing the parameters from a visually similar project. The learned parameters are usually located in the "IoU" pad in the output directory as a JSON file.

5. Method Selection: This checkbox menu lets you choose the segmentation method. The GUI includes various methods like Sauvola, Niblack and Gaussian. Notably, Sauvola method is considered the best choice due to its effectiveness and robustness in segmentation of spheroids on a variable background subject to inhomogeneous light conditions.

6. Gradient Descent Parameters: These parameters are crucial for the Gradient Descent algorithm. They include settings like learning rate, number of iterations, and batch size. It's generally recommended not to alter these parameters unless you have specific requirements or understanding of the algorithm. Still, a low batch size value (ranging from 4 to 16) is recommended for computing devices with low RAM capacity.

7. Hole Finding Settings: In this section you can set the mode of searching for holes inside spheroids. If it is suspected that some of the spheroids to be segmented contain holes, the "Detect all contours" checkbox can be checked. This selected mode also searches for internal contours (holes) inside the spheroids. In the resulting segmented image in the subfolder segmented_images/.../results, it marks them in blue, while the outer contours are marked in red. If some images contain inner contours and some do not and the user is not sure which option to choose, the last checkbox "View results for 'all holes' and 'no holes' and select" can be checked. This option displays a selection dialog where it is possible to choose for each which of the two options mentioned above the user finds better and save that one.

8. Other Settings Checkboxes: This section contains various checkboxes for additional settings. These settings might include options for finding also inner contours, the possibility to create a zip file in COCO 1.0 format, which will contain the resulting segmentations and can then be uploaded to the CVAT platform. It is also possible to calculate the characteristics and properties from the obtained outer contours, whose values are then stored in a separate Excel file.

Additional Information

• The GUI is structured to facilitate both beginners and advanced users in the field of image processing.
• Each feature and button is designed to provide maximum control over the segmentation process, ensuring that users can tailor the process to their specific data and requirements.
• Regular updates and feedback from users are encouraged to continuously improve the functionality and user experience.

Using the Spheroid Quantification GUI

The Spheroid Quantification GUI offers to determine the characteristics of the spheroids based on the provided binary images of the spheroid masks. These characteristics are, for example, spheroid area, circumference, compactness, sphericity, diameter and others. The output is then saved as an XLSX (Excel) file in a user-defined folder.

Support

For detailed explanations of the segmentation methods and Gradient Descent algorithm, refer to the technical documentation provided with the package at Github prusek-spheroid.

If you encounter any issues or need further assistance, please feel free to reach out to [e-mail]

For any issues or queries, contact me at: [e-mail]